Three years ago, Alibaba was competing with every other major tech company to secure Nvidia H100 chips. Now it's building its own.
Alibaba announced a new in-house AI chip alongside a new model, part of a push toward what the company calls a "full-stack" AI strategy - meaning it wants to own the entire chain from the processors training the AI down to the models running on top of them.
The strategic backdrop is US export controls. Since 2022, the US has progressively restricted sales of advanced AI chips to China, blocking Nvidia's most powerful data center processors from the Chinese market. Chinese companies that once relied on Nvidia for training large models - the computationally intensive process of teaching AI on massive datasets - now face a choice: pay a premium for restricted alternatives, use less capable hardware, or build their own. Huawei went down the build-your-own path with its Ascend chip line. Alibaba is now doing the same.
The logic for in-house chips goes beyond supply security. A company that controls its own silicon can design it specifically around how its own models run, rather than adapting software to general-purpose chips. That's a meaningful efficiency advantage - similar to how Apple's M-series processors outperform equivalent Intel chips on Apple's own software because the hardware and software were designed together.
Alibaba already operates one of China's largest cloud platforms. If its own chips prove competitive, that reduces costs and eliminates the risk of further supply disruption from future US policy changes. It also positions Alibaba to compete with cloud providers that sell AI compute as a service - a market worth tens of billions annually.
The missing details are the ones that actually matter: how this chip performs against Nvidia's H800 (the restricted, lower-spec version currently available in China), which model workloads it supports, and whether it's production-ready or a research demonstration. Those numbers will determine whether Alibaba has built a real alternative or a strategic announcement that still runs on Nvidia hardware when it counts.